Section: New Results

Resource Allocation in Large Data Centres

Participants : Christine Fricker, Philippe Robert, Guilherme Thompson.

With the exponential increase in internet data transmission volume over the past years, efficient bandwidth allocation in large data centres has become crucial. Illustrating examples are the rapid spread of cloud computing technology, as well as the growth of the demand for video streaming, both of which were quasi non-existent 10 years ago.

Currently, most systems operate under decentralised policies due to the complexity of managing data exchange on large scales. In such systems, customer demands are served respecting their initial service requirements (a certain video quality, amount of memory or processing power etc.) until the system reaches saturation, which then leads to the blockage of subsequent customer demands. Strategies that rely on the scheduling of tasks are often not suitable to address this load balancing problem as the users expect instantaneous service usage in real time applications, such as video transmission and elastic computation. Our research goal is to understand and redesign its algorithms in order to develop decentralised policies that can improve global performance using local instantaneous information. This research is made in collaboration with Fabrice Guillemin, from Orange Labs.

In a first approach to this problem, we examined offloading schemes in fog computing context, where one data centres are installed at the edge of the network. We analyse the case with one data centre close to user which is backed up by a central (bigger) data centre. When a request arrives at an overloaded data centre, it is forwarded to the other data centre with a given probability, in order to help coping with saturation and reducing the rejection of requests. In [16] , we have been able to show that the performance of such a system can be expressed in terms of the invariant distribution of a random walk in the quarter plane. As a consequence we have been able to assess the behaviour and performance of these systems, proving the effectiveness of such an offloading arrangement.

In a second step, we investigated allocation schemes which consist in reducing the bandwidth of arriving requests to a minimal value when the system is close to saturation. We analysed the effectiveness of such a downgrading policy, which, if the system is correctly designed, will reduce the fraction of rejected transmissions. We developed a mathematical model which allows us to predict system behaviour under such a policy and calculate the ideal threshold (in the same scale as the resource) after which downgrading should be initiated, given system parameters. We proved the existence of a unique equilibrium point, around which we have been able to determine the probability of the system being above or under the threshold. We found that system blockage can be almost surely eliminated. This policy finds a natural application in the context of video streaming services and other real time applications, such as MPEG-DASH. A document is being written to further publication.

Finally, with those results, we now try to extend our research towards more complex systems, investigating the behaviour of multiple resource systems (such as a Cloud environment, where computational power is provided using unities of CPU and GB of RAM) and other offloading schemes, such as the compulsory forwarding of a request when it's blocked at the edge server, but keeping a trunk reservation to protect the service originally assigned to the big data centre.